Evolutionary design of collective communication based on prediction of conflicts in interconnection networks

collective communications, evolutionary optimization, interconnection network, multiprocessor

This work describes the application of a hybrid evolutionary algorithm to scheduling collective communications on the interconnection networks of parallel computers. To avoid contention for links and associated delays, collective communications proceed in synchronized steps. The minimum number of steps is sought for any given network topology, store-and-forward switching, minimum routing and given sets of sender and/or receiver nodes. Used algorithm is able not only to re-invent optimum schedules for known symmetric topologies such as hyper-cubes, but it can find schedules even for  asymmetric or irregular topologies in case of general many-to-many collective communications. In most cases the number of steps reaches the theoretical lower bound for the given type of collective communication; if it does not, non-minimum routing can provide further improvement. Optimum schedules are destined for writing high-performance communication routines for application-specific networks on chip or communication libraries for general-purpose interconnection networks.